A Solution of Short Term Unit Commitment in Smart Grid Using Artificial Neural Network and Fuzzy Logic

Abstract

The unit commitment problem is proposed to be solved by the Artificial Neural Network and fuzzy systems in smart grid using distributed energy storage systems. The main objective of short term load forecasting is to provide load predictions for generation scheduling, economic load dispatch and security assessment at any time. With the increased attraction of medium and smaller scale systems the expansion of the integration of intermittent renewable resources in smart grids. The effectiveness of the economic dispatch is well under stood where the objective is to schedule the committed generators to meet the load. This paper is intended to offer a tool for analyzing potential advantages of distributed energy storages in Smart Grids with reference to both different possible renewable resources at the distributed energy environment. The objective of this paper is to commit the units with the AI techniques back propagation neural network and Fuzzy logic without violating the constraints and to compare the results in order to find the best suitable method. Index Terms: Unit commitment, Smart grids, Distributed energy storage, network, Artificial Neural Network and Fuzzy System . I. Introduction Among these innovative alternatives, Distributed Energy Storage Systems (DESS) is nothing but the grid offer interesting features and have received considerable attention lastly. However, they still remain capital- intensive and raise cost-competitiveness issues. This paper deals with the potential of DESS to support and to optimize distribution system operation. First of all, some general information about DESS and their technical performances is given. Then, an overview of their applications in liberalized power systems is drawn (for DSO and other stakeholders). In the last part, a new approach to the combination of storage benefits is introduced and its use within the framework of this project is briefly discussed. In any power system, the load is dynamic in nature. It is higher during the daytime and early evening when industrial loads are high; lights are on, and so for the and lower during the late evening and early morning when most of the population is asleep. The load variation is continuous and the load must be met with the available resources economically. This is done by committing (switching ON) and decommitting (switching OFF) of the units in power station. By running only the most economic units, the load can be supplied to the best efficiency of unit operators. Thus committing the correct number and kind of units such that the load is met at least operating cost. There have been many methods that are available to solve the unit commitment problem such that the Lagrangian method, Dynamic Programming (DP), branch and bound technique, simulated annealing, Priority listing and Advanced Priority listing method. The DP method based on priority list is flexible, but it computational time suffers from dimensionality. Lagrangian relaxation for UCP is superior to DP due to its higher solution quality and faster computational time. However, numerical convergence and solution quality of LR do not give satisfactory results in case identical units exit. These methods though may give results but does not give a qualitative interpretation of the results in terms of input variables. The neural network computing enhanced by expert systems has opened up a new route for the optimization of generation scheduling. With proper and sufficient training, the information regarding the optimal operation of a system can be stored in the network as such, and the output can be obtained in a much shorter time. In the problem, multilayer feed forward network using back propagation error of learning determine variables corresponding to the operating level of generators and production cost. Load demand profile is input to the neurons in the input layer. Generation is the output to the neurons in the output layer. The fuzzy system involves the fuzzification and defuzzification process with the certain membership functions to obtain the predicted load demand.